Robot Talk Season 2, Episode 1 (September 2021): Swarm Robotics

Swarm robotics is when many robots come together to work on a common goal. Researchers working to develop these self-organising, multi-robot systems often find inspiration within the natural world, and there are certainly many striking parallels between the behaviour of robot swarms and living organisms – from mimicking coordinating flocks of birds that can travel long distances and seek safety in numbers, to harnessing the example of industrious colonies of ants, bees, wasps and termites, which gain many benefits through division of labour and cooperation using comparatively simple rules.

Dr Claire Asher of the EPSRC UK-RAS Network recently interviewed two experts in this fascinating area of robotics research for the Season 2 launch episode of our popular Robot Talk podcast. Claire first spoke with Merihan Alhafnawi, PhD researcher at the University of Bristol working on expressive robot swarms.

CA: So, Merihan, how many robots constitute a swarm?

MA: It hasn’t really been defined – it depends on the application and what we’re using the robots for and where, so it can be anywhere from tens of robots to thousands of robots. We do know that just two robots does not constitute a swarm! 

CA: Your PhD is focusing on expressive robot swarms. What do we mean when we say expressive in the context of robots?

MA: When we have a swarm of robots, it’s important that we know how we can communicate with the swarm if we want to tell it something.  So imagine, for example, the situation where there is a change of requirements within a stated mission, or if the robots are not behaving as you’d expect them to, you’d want to communicate with them.  In order to have this communication, you need the swarm to be expressive, so that you can understand the state of the swarm. “Feedback” is also really important – once you’d told the swarm something, you want to be sure that the swarm has understood what you’re trying to say. So the swarm, as a collective, needs to be able to express itself to you, so that you can confirm that your input has been received and understood.

CA: So the idea is to make the interactions between swarm robots and humans more fluid?

MA: That’s true. It’s not very intuitive how one interacts with many robots – which could number thousands – and it would be very cumbersome to interact with each robot singly. So it’s a question of optimising how you, as a human operator, interact with the swarm as a single unit.

CA: What kind of uses and environments are swarm robots applied to?

MA: Swarms have been widely studied in many applications. Many of the applications are still studied in the lab, because we’re at the early stages, but a lot has been researched in fire-fighting, where a swarm of many drones are sent to try to fight forest fires. Other applications include using robots to run autonomous warehouses, which frees up human resources to do other tasks that are more important. Other key applications are planetary exploration – sending a swarm of robots to explore another planet – and search and rescue, where a swarm of robots are sent to look for survivors in areas that might challenging or dangerous for humans. In my research right now we’re thinking of a new social context in which we can use robots, which is using swarms to help with tasks such as collective decision-making and brainstorming in meetings.  For this, we have robots with screens that we use as Smart Post-It notes. The idea is that robots self-organise and aggregate themselves based on themes, so it’s easier for humans to quickly go through and look at ideas.

CA: In terms of modes of expression for expressive robots, are these typically audio or visual?

MA: That’s a really good question, which depends on the hardware that you’re using. For the swarm that we’re building for a social context, for example, we have a touchscreen to provide visual feedback for humans. If the swarm is made up of drones fire-fighting in the field, you wouldn’t be able to see them, so we’ve been researching the best ways to propagate messages. This could be talking to the closest drone to you, and the drone would then spread the message to its peers. The message would then be propagate back to you, to deliver a status report. Human-swarm interaction methods depend greatly on the context and the hardware used, which can be challenging.]

CA: When we talk about this interaction, I immediately think about the waggle dance in bees!

MA: Absolutely. Swarms are greatly inspired by nature, and bees are a huge inspiration. There are algorithms that are inspired by bees’ behaviour in nature, and similarly by flocks of birds and schools of fish.

CA: During the UK Festival of Robots in June, you ran a great demo of your robots that virtual participants could take part in.  How do you find members of the public reacted to swarm robots?

MA: The most interesting part of public engagement is to see how people react to the research, and responses vary greatly. Kids are mostly always intrigued by robots, and are always pushing their parents to come and talk to us.  They are very drawn towards the robots – touching them and moving them around – so that’s a great segue to talking about robotics and STEM. The conversations we have with kids are usually really interesting. With adults, it greatly differs. Most of the time we get really positive feedback – that the work we’re doing is great and interesting, and that they’re happy that this kind of research is taking place in the UK. They also sometimes come up with really interesting ideas about where you can take the research. We do sometimes see people who are more sceptical and ask if robots will replace jobs, and this is interesting, because it clears the way to explore what we’re intending to do.  Recently we conducted an experiment in Cabot Circus, one of the largest malls in Bristol, where we took a swarm of robots and asked the public questions about the most impactful actions that we should take to battle climate change (recycling, eating less meat, using more energy-efficient appliances and so on) and we had a great response to this.

CA: Swarm robots in popular culture are quite often depicted as a force for bad, and the idea of robots acting on their own as a swarm and in a coordinated way might inspire a little bit of fear. So is it important to go out and break those misconceptions, and also make kids aware of robotics as a career option?

MA: Absolutely. As part of the UK Festival of Robotics, we had an outreach activity to introduce kids to robotics. When I was a kid, the focus was on computers, and now it’s about robotics, so it’s really nice to see the evolution of these outreach programmes.

For more information about the University of Bristol’s research into swarm robotics and swarm intelligence – including decision making, exploration, collective motion, shape formation and mapping – please visit: https://www.bristolroboticslab.com/swarm-robotics

To talk more about how swarm robots is being inspired by nature, Claire spoke with Dr John Oyekan, a lecturer in Digital Manufacturing at the University of Sheffield, whose work on swarm algorithms has been applied to the automotive, aerospace and manufacturing sectors.

CA: My PhD was on ants, looking at division of labour and colony organisation. I’m really interested to know how much has swarm robotics been influenced by the way that colony-living insects like ants actually behave in nature?

JO: There is quite a lot of inspiration from ants and bees, and we try to learn from these organisms and develop similarities in robotics. One of the reasons people look at biological organisms is that there is a lot to learn from them. From a simplistic view, they are many (in number), and they use their collective mechanisms to perform tasks that as an individual they can’t do on their own. That has inspired engineers to learn from them to try to replicate the same thing in technology.

CA: Does the robot swarm have a queen?

JO: In nature, when the swarm is actively operating, they don’t have anyone calling the shots. There are queen bees, but they are in the hive while the workers are outside working. So how do they collaborate and coordinate their actions? That question (is what) really interests scientists and engineers.

CA: We want swarm robots to interface with human controllers, but that can be challenging enough with one robot. How do you manage the interface between humans and many robots?

JO: That’s a great question. An individual human controlling a single robot is difficult, so when you add (more) numbers, humans start to experience cognitive load distress because it’s difficult to keep track (of each agent). That’s why it’s very important that the agents themselves have a level of autonomy to enable them to act locally but also try to effect a global task. My view is that the best way to manage this interface is by giving the swarm a global goal to achieve and leave it to the swarm to find a way to achieve it.

CA: So the high level commands come from a human and the robots figure out how to make that happen?

JO: Exactly. Humans stay at a supervisory level and (human-swarm) collaboration takes place at a high level.  So, for example, you might instruct a swarm to carry an item.  So the robots fly to the item and say “what is the best way to configure ourselves to pick it up? How do we place ourselves optimally to carry the load?”. They can then use their sensors to sense the object and figure out how to complete the task autonomously.

CA: Presumably there has to be some kind of fail-safe where the robots can report back if there’s an obstacle, or if they can’t complete the task?

JO: You always make sure that the human controller is in the loop of the system, because you don’t want your swarm robots to do something you don’t want them to do. It’s an interesting concept, meaning that the human might need to break down the task into smaller goals to enable the swarm to use the level of intelligence they have to walk through a series of sub-goals on order to achieve the larger goal.

CA: How hard is to stop drones from bumping into other? Birds use a set of relatively simple rules to ensure that they don’t collide – is it the same for robots?

JO: It’s tougher for robots. With Unmanned Aerial Vehicle (UAV)s, you have to do sensing and you then have to process that data. Depending on the sensing modality you use, the amount of data you get from that sensor could be huge. If you’re trying to put all those computational resources on a single UAV, it becomes very difficult to crunch through that data and use the outcome to exert control.  The feedback loop that needs to take place can be characterised as Sense-Think-Control. This loop has to complete as fast as possible to ensure that the agents are able to sense their environment and avoid collisions with obstacles, and avoid collisions with each other. That is what makes it challenging. How do you ensure that the time between sensing and control is as short as possible, to ensure that you are able to respond quickly in the environment? It’s challenging, but it also raises some interesting research questions. Similarly, how do you reduce the amount of data that you need for the feedback loop?  If you look at academic literature, there are various ways of doing that, and it’s still a research question in itself. The biggest research question, I think, is how you get a swarm of UAV to flock like starlings?  One approach is to use GPS signals, like at the Olympics, whereby you know the precise position of the agents. Biological organisms, you could argue, only use vision to sense and act. That is what makes swarm research so interesting – how to achieve that level of biological accuracy.

CA: Some of your work is in using swarm algorithms for search-and-rescue. I was intrigued by the idea of visualising an invisible pollutant.

JO: The idea for this came about because I was reading about the idea of invisible pollutants. This could come, for example, from terrorist attacks, where Sarin gas is released into the environment, or even natural disasters where invisible carbon dioxide is released from the underground reservoir of a lake, as happened in Cameroon in 1986. That got me thinking: how do we get people to see where the dangerous substance is and run away from it?  So, we could use a swarm of robots that can fly and move about in the environment, and when they find the pollutant, they are able to arrange themselves to fit the profile of the substance. Because you can see the robots, an invisible problem becomes visible. As the hazardous substance moves through the environment, the robots can configure themselves and people can see where the substance is and keep themselves safe and away from those areas.

CA: That’s such a clever idea. It terms of how the robots know where the hazardous substance is, presumably they need chemical sensors?

JO: Absolutely. The chemical sensors have to be suited to the hazardoussubstance they’re trying to sense, because different chemical sensors require different ways of sensing the modality in the environment. You need to know what specific application you’re deploying the swarm into, and equip them well for that task.  Looking at it from a biological perspective, ants use pheromones that have a particular message they’re trying to communicate, so each ant’s chemical receptor is designed to sense all those different chemical modalities and respond as appropriate.

Find out more about the work of University of Sheffield’s Department of Automatic Control and Systems Engineering: https://www.sheffield.ac.uk/acse

To explore the research groups in the University of Sheffield’s Department of Automatic Control and Systems Engineering, please visit: https://www.sheffield.ac.uk/acse/research/groups

To explore the University of Sheffield’s investigations into robotic systems inspired by nature and robotic models of natural systems, please visit: https://www.sheffield.ac.uk/naturalrobotics

To listen to Robot Talk Episode 10 – Swarm Robotics, click ‘play’ below.

Robot Talk is available on all major podcast providers, including: